What check.py refuses

Generated from check.py itself. Every line below is a message the checker prints when something is wrong — not a paraphrase of one. There are 314 guarantees in 143 groups, and every one of them is exercised by a mutation in test-check.py, which says so on every run.

A contributor should not have to read 4822 lines of Python to find out what will fail. That figure said eleven hundred until September 6, 2026, on the page whose whole reason for existing is that the Python is too long to read — it was written when the checker was a quarter of its size and nothing recomputed it. The regenerator writes it now.

It is not checked, and that is worth a line. A guarantee inside check.py that the file is N lines long cannot hold: the mutation harness instruments the checker by rewriting it, which makes it three lines longer, so the check would fail on every run of the test that exists to prove the checks work. A number that describes a file cannot be verified from inside that file when the tool verifying it edits the file. Regenerated, not guarded. This page is that list, and check.py regenerates it and refuses to pass if it has drifted — so it cannot fall behind the checker it describes.

The … marks where a message fills in a word, a count or a sentence.

dictionary

• …: …
• …: no etymology
• … roots are shorter than four letters; the short space is closed at 49 — see CONTRIBUTING rule 2
• ACCEPTED_PAIRS lists …/…, which is no longer a pair in the dictionary
• duplicate entries: …
• new minimal pair: … / …
• l/r minimal pair: … / … — the one contrast the language forbids

phonology.md is a record

• grammar/phonology.md does not list the … ’…’ (…)
• grammar/phonology.md lists the … ’…’, which no root contains

texts

• …: no code block — a text keeps its Amadunia inside one, and nothing below can read it
• …: ’…’ is not in the dictionary
• …: no checkable ’## Roots used’ section
• …: claims … roots, uses …

lessons

• …: lesson number must be two digits

indexes

• index-english lists non-words: …
• index-english is missing: …
• grammar/README.md does not link … from its table
• lessons/README.md does not link …
• …: no ’## Open questions’ section — a rule records what it left open, even if that is nothing
• grammar/README.md lists … open questions under …; ….md has …
• grammar/README.md says … open questions; the files have …

a decided briefing is over

• …: a live open question points at …, which says it was decided — strike the question through or the briefing is not over
• …: calls … ’…’ — that briefing has been decided

copula

• …: ‘es …’ — es never goes before an adjective: …
• …: ’… …’ — no plural after a number: …
• …: ’… no’ — no comes before the tense particle: …
• …: ’… …’ — the owner goes before ini/itu: …

lesson order

• README.md: no ’## Learn the basics’ section — that is where the front page teaches its words, and the lesson order is measured from it
• …: uses ’…’ before any lesson teaches it: …
• …: claims … roots taught by here; the lessons have taught …
• …: prerequisite should be Lesson …

one root, one introduction

• …

coverage

• …

the syllabus may not credit a late lesson

• lessons/README.md credits Lesson … with the rising-tone question; Lesson … already asks one, and that lesson page says so itself

no chains

• …: verb chain ’… …’ predates Lesson …

verb position

• …: ’… …’ — … is not a verb: …

the other copula

• …: ’… …’ — a noun predicate needs es before it: …

a gap claim must cite a page

• …: a gap claim cites no page — …

a lesson may not teach an open form silently

• …: uses daima or kadang, whose position is an open question, without linking the briefing that says so

a noun after ini closes nothing

• …: ’… …’ — ini and itu close a phrase, so a noun after one is a predicate and needs es: …

dictionary order

• dictionary.md: the … group is not in numerical order
• …

counts named in a link

• …: ’…’ names … open questions; there are …

shown, not just glossed

• … roots appear in a lesson word table and in no lesson sentence: …

two roots joined

• …: ’…’ joins two roots and is neither a number nor a plural — see grammar/word-formation.md: …

an etymology names a language

• dictionary.md: ’…’ names no language in its sources (…); ‘already-global’ is a label, not an etymology

no articles

• dictionary.md: ’…’ is glossed as an article (…); the language has none — see grammar/definiteness.md

the names briefing counts itself

• proposal-names.md says … sentences are formally ambiguous; the corpus has …
• the ambiguous-name count is stale where it is written: … against a recount of …

the adjective-fragment count counts itself

• … says … two-word utterances are a noun with an adjective; the corpus has …

the modal-adjective briefing counts itself

• proposal-modal-adjective.md says … blocked combinations; three modals against … adjectives is …

every rule says where it is taught

• …: does not say which lesson teaches it

frequency

• frequency.md’s total is stale; the corpus has … running words
• frequency.md is missing or contradicts the row ’…’
• …
• frequency.md’s top forty has drifted: …

es before a preposition

• …: ‘es …’ — es stands before a noun and before nothing else: …

a word nobody invented

• … line …: ’…’ is not a word in the dictionary: …

the adverb keeps its verb

• …: ’…’ is glossed as an adverb but … stands after the object, where it describes the object: …

an adjective’s English name

• …: the gloss says ’…’, which is …, and no such word is in the Amadunia: …

reading ladder

• reading-ladder.md is missing or contradicts the row ’…’

stress

• stress.md’s syllable counts are stale; the dictionary gives …/…/…
• stress.md’s agreement figure is stale; recount gives … of …, …%

the poem still scans

• texts/text-5-uan.md: no code block, so the poem cannot be scanned
• text-5’s closing line no longer scans as the page says: … is …
• …: odd number of code fences — one is unclosed

stress marks are right

• …: ’…’ marks the wrong syllable; the beat is on ’…’

the question word keeps its place

• …: ‘ke’ opens the sentence — the question word stands where the answer will stand, so it follows es: …

the ladder’s number is the binding one

• reading-ladder.md is missing or contradicts the row ’…’
• texts/README.md’s index does not list …

a lesson may not use a rule dated after itself

• …: uses the … rule, which Lesson … introduces: …

giving needs por

• …: ’… …’ after beri — a recipient takes por, or it reads as an owner: …

the number comes first

• …: ’… …’ — the number goes before its noun: …

the adjective follows

• …: ’… …’ — the adjective goes after its noun: …

possession order, preposition

• …: ’… …’ — a preposition takes a noun: …
• …: ’… … …’ — the owner comes before the adjective: …

numbers parse

• the dictionary gives … roots glossed as a bare number; the number system is ten digits and the bases sen and mila
• …: the number system is … words, … of which reach a hundred; the page does not say ’…’
• numbers.md: the largest settled number is … and the status line does not name it
• dictionary.md: the count the built numbers are excluded from is not …
• …: the digit table does not match the dictionary, which gives …
• …: ’…’ does not build a number — …

the guarantee list

• GUARANTEES.md has drifted from check.py — regenerate it
• GUARANTEES.md’s counts are stale; check.py has … guarantees in … groups

place before time

• …: time comes before place, and the order is place then time: …

the joined-form count

• grammar/word-formation.md: … joined forms are neither a number nor a plural
• word-formation.md’s count is stale; the repository has … joined forms
• word-formation.md’s breakdown is stale; … numbers and … reduplications

nothing is orphaned

• …: nothing links to it — no path from README.md reaches it

tables render

• …:…: table rows with no header row above them — …

root in use

• …

the A1 checklist

• proposal-a2.md’s checklist row for ’…’ is stale; the dictionary has … of …
• proposal-a2.md’s missing list for ’…’ has drifted; the dictionary gives: …
• proposal-a2.md: says ’… gaps found by writing’; the table holds …
• proposal-a2.md: the checklist is missing … and … of those were found by writing, so it adds … more
• proposal-a2.md: asks which of the wrong number matter; the checklist is missing …
• proposal-a2.md: … concepts are missing and the page must say so in words as well as in figures
• proposal-a2.md’s checklist total is stale; recount gives … of …
• proposal-a2.md says the first … and then lists …
• proposal-a2.md’s section 3 does not divide the gap table as it stands: … gaps, … found by writing, … by a sweep, … by a question
• proposal-a2.md: kinship’s absences are … of …
• proposal-a2.md: kinship is … of … present

how many sentence shapes

• texts/README.md’s shape count is stale; the material has … sentences in … distinct shapes
• texts/README.md’s repetition figure is stale; … of … sentences are distinct
• texts/README.md’s most-repeated sentence is stale; it is ’…’, … times across … files

a text’s table repeats its own text

• …: the line-by-line table has a sentence the text does not: …

how many rules, how many briefings

• README.md’s directory table does not say ’… rules’; grammar/ holds that many rule pages
• proposal-modal-adjective.md does not say there are now … briefings; grammar/ holds that many
• README.md’s directory table does not say ’… briefings’; grammar/ holds that many
• grammar/README.md does not say there are … briefings
• grammar/README.md does not say … briefings are open

every rule is exercised, not just used

• texts/README.md’s rule table has drifted from the scan that produces it — regenerate it
• the rules the ladder dates and the rules the scan finds differ: dated but never found …, found but never dated …
• texts/README.md: the material has … questions and … of them use no question word
• texts/README.md claims a floor of ’…’ texts per rule; the thinnest is … in …

a gloss may not claim a gap

• …: a conversation is translated with ’…’, which the language does not have or has not settled
• …: the gloss says ’…’, and … is on the list of words the language does not have: …

one root, one job

• dictionary.md: ’…’ is glossed as both a noun and a verb (…) — one root, one job, see CONTRIBUTING rule 8

the front page teaches every rule

• … is settled and the front page has no row for it — add one and name it in check.py’s _FRONT_LABEL
• README.md no longer teaches …: ’…’ is gone from the page that says it is all of the grammar
• …: says ‘all …’ hyphenated forms; the repository has …

the page that claims to be the whole language

• … is settled and check.py cannot tell whether Lesson 23 names it — add it to _L23_NAMES
• lesson-23-everything-so-far.md says it is the entire language and never names …

a lesson may not send a learner to a text they cannot read

• …: tells the reader they can read …, which reading-ladder.md opens at Lesson … — … lessons later

madad is still held back

• madad is used in … sentence(s) — … — which answers the open question by use; see proposal-two-jobs.md

the frequency briefing counts itself

• proposal-frequency.md’s row for ’…’ is stale; the corpus has …
• proposal-frequency.md’s tena figure is stale; recount gives … of …
• …: says tena stands last in … of …; the recount gives … of …
• proposal-frequency.md: answer … costs … of … rewrites, because … sentences already stand there
• proposal-frequency.md’s headline is stale; … of … sentences stand in the adverb slot

an unmarked verb glossed as past

• …: ’…’ has no suda and is glossed in the past (’…’) with nothing before it setting the time: …

a negative sentence glossed without one

• …: ’…’ denies something and its translation does not: …

a place word must end its clause

• …: ’… …’ — place goes last, and a place word takes no noun after it: …

a text’s own word count is a claim

• …: says … words; its text has …

an adjective is not a noun

• …: ’…’ is an adjective standing as a subject; the language has no nominalisation: …

a number written in Amadunia is a claim

• …: the numeral in ’…’ is …, which its translation does not say: …
• …: says Amadunia has … words; the dictionary has …

no denies a predicate, not a noun

• …: ’…’ denies a noun with no verb of its own — no goes before a predicate, and the language has no ellipsis: …

every sentence needs a predicate

• …: ’…’ is translated ’…’ and has no predicate in it — no verb, no adjective, no place word and no es

ta is he and she, and the pages must show it

• …: glosses ta … times and always as …; ta is he, she and it
• pronouns.md’s gender figures are stale; the corpus gives … sentences, … gendered, … she, … he, … both, … it

counts that sit outside the link

• …: ’…’ stands in a sentence about the open questions and is none of … questions, … briefings or … open: …
• …: ’…’ — there are … open briefings and … open questions
• …: says ’…’; texts/ holds …

a text counting its own sentences

• …: says ’…’; its text has …
• …: heads its summary ’…’; its text has … sentences
• …: says ’…’; its text has … sentences
• …: says ’…’; its text has …
• …: says ’…’ opens … sentences; it opens …

a text counting the texts that leave mi out

• …: says ’…’; … do — …

a text counting its own commands

• …: splits into … sentences but has … stops — the two readings disagree
• …: says … of its sentences are commands and … of those negative; the text has … and …

how many texts there are

• …: says ’…’; texts/ holds …

the prose that reads the reading ladder

• reading-ladder.md’s list of lessons that move nothing is stale; the table gives …: …
• reading-ladder.md: Lessons 24 and 25 carry …% to …%
• …: ’…’ — the ladder first reaches …% at Lesson …

the gap list counts its own two methods

• dictionary/README.md: … rows name a text or a lesson, and the page does not say so
• dictionary/README.md: the … gaps found by a question are not named as numbers … to …
• dictionary/README.md: the three parts of the gap list are … and must account for all … of it
• dictionary/README.md: the feelings count is not stated against … gaps
• dictionary/README.md: a gap row has … cells against the header’s … — …

the demand table counts the pages that ask

• grammar/README.md’s demand row for ’…’ says …; … pages write the claim
• grammar/README.md’s demand table is not in descending order of demand, which is what it says it is for: …
• …: says ’…’; the demand table has … rows at zero

a lesson heading counting its own verbs

• …: says ’…’ and its word table teaches …

the two-jobs briefing counts rabota

• proposal-two-jobs.md’s heading is stale; rabota stands in … places in the material
• …: says rabota stands in …; the material gives … places
• proposal-two-jobs.md’s split is stale; … places put rabota after a subject and … put something in front
• proposal-two-jobs.md’s distinct-sentence figures are stale; the material gives … and …, … in all

the two-job problem has two roots, not one

• …: raises madad’s two jobs without linking the briefing; the problem covers two roots and that page separates them
• …: calls a root the only one holding two jobs; the audit that said so read the glosses and not the sentences, and proposal-two-jobs.md separates two

a comma list still needs its aur

• …: ’…’ is a list of three or more with no aur before the last — commas stand in for all but that one

a question word as a noun predicate takes es

• …: ’…’ answers with a noun, so the question word is a noun predicate and takes es

a lesson may not call a settled rule undecided

• …: ’…’ — … is settled

the ordinals figure is on four pages

• …: says ordinals would remove the need for ’…’ roots; it is nineteen — seven weekdays and twelve months

the modal-adjective briefing counts the material

• proposal-modal-adjective.md’s row ’…’ is stale; the material gives …
• …: says … verbless adjective predicates; the material gives …
• …: says … working modals; the material gives …

no one sentence holds every rule of the language

• …: ’…’ — nine of the … rules are in it, and no sentence can hold a command and a question at once

a rule page may not say nothing has asked for it

• plural.md says nothing has tried to say all or some; … page… record wanting every

a word a rule page names is not a noun

• …: the heading ’…’ names … as a rule word and check.py classes it as a noun

two degree words in one slot

• …: ’… …’ — two degree words in one slot, which comparison.md records as not granted: …

a rejection that cites a minimal pair must be one

• …: rejects … as a minimal pair with …, and they differ in length or by more than one sound
• …: … is a minimal pair with …, and the table says …

“one sound from” has to be one sound

• …: says … is one sound from …; it is …

how many nouns are glossed both ways

• definiteness.md says … nouns are translated both ways; the material gives …

a page may not describe a correction it caused as pending

• …: ’…’ on compounding, and README.md has already withdrawn the claim

how many sentences the existential stands in

• sentence-types.md’s existential count is stale; a sentence-initial es stands in … sentences

the but briefing cites real pages

• proposal-but.md says ’… pages’ but its evidence table has … rows

the but briefing’s prose counts its own table

• proposal-but.md’s prose counts ’…’ pages; its table has …
• proposal-but.md does not call page … the one that matters most
• proposal-but.md cites …, which does not exist
• proposal-but.md quotes ’…’ from …, which no longer contains it

used by a text, not a lesson

• proposal-a2.md says ’… are still unused’; … roots appear in no text
• proposal-a2.md’s heading says ’… roots have never been used’; … do
• … roots are in no text and nothing forbids writing them: …

derived documents

• texts/README.md lists …, which does not exist
• texts/README.md counts …, which has no code block
• texts/README.md says … uses … roots; it uses …
• index-english.md is missing the headword ’…’ (…)
• index-english.md has a headword the dictionary no longer gives: ’…’
• index-english.md maps ’…’ to …; the dictionary now gives …

settled, not open

• …: lists ’…’ as still open; grammar/ marks it settled

closed gaps

• …: says there is no word for ”…”, but the dictionary now gives …

machine-readable

• dictionary.json has drifted from dictionary.md — regenerate it
• dictionary.json could not be read: …
• dictionary.csv has CRLF line endings; write it with lineterminator=‘\n’
• dictionary.csv has drifted from dictionary.md — regenerate it

balance

• balance.md counts … roots; the dictionary has …
• balance.md’s European figure is stale; recount gives … of …, …%
• balance.md’s list of families named in a fifth or more is stale; the dictionary gives …: …
• …
• …

the A2 briefing’s root-length table

• proposal-a2.md’s root-length table has drifted from the dictionary — regenerate it
• …
• balance.md’s largest family is stale; recount gives … at …%
• stress.md’s cost figures are stale; recount gives … keeping the beat and … losing it
• stress.md’s share figure is stale; … of … is … per cent
• stress.md’s ratio is stale; … to … is …
• stress.md: the dictionary has … roots of three syllables
• stress.md: the two sets of … share … roots — … — and the page must name them
• stress.md: … of the moved roots have two syllables, not what the page says

conjunction.md’s rejected candidates

• questions.md rejects … for containing …, which is not in it
• conjunction.md: … roots end in -…

every ordering in this file is fully decided

• check.py sorts with a single-term key — … — so ties fall in the order a set happened to iterate in

the rule detectors are defined exactly once

• check.py detects the … rule in more than one place; the copies drifted three ways before they were merged

how many rules the exercise table can see, and cannot

• texts/README.md: grammar/ holds … rule pages
• texts/README.md: the scan tracks … rules

how many roots the two thin families have actually given

• proposal-a2.md: … roots come from Chinese or Japanese

a rejection that counts syllables or letters must count

• …: says … has … …; it has …

every “N roots end in -X” claim, everywhere

• …: says … roots end in -…; … do

how much of the corpus count is nobody’s sentence

• … runs come from explanatory prose, and both pages that state it must say so

how many roots reach a sentence of more than one

• frequency.md: … of … roots appear in a sentence of more than one word
• frequency.md must name the … roots that reach no such sentence: …

lesson 26 counts the turns in its conversation

• lesson 26 names … commands and each must begin with a verb: …
• lesson 26’s conversation has … sentences
• lesson 26: counted strictly there are … one-word turns and … of the two kinds together

text 21 measures itself against the others

• text-21 uses … distinct shapes by the scanner that produces the corpus figure
• text-21: the next highest shape count is …
• text-21: its three commonest shapes cover …%
• text-21 is the …th least repetitive of … by its three commonest shapes
• text-21 shares a shape between …% of adjacent pairs, …th of …

text 5’s arithmetic about rhyme

• text-5: … of … roots end in a vowel, …%
• text-5: …% of roots end in -a
• text-5: -ar and -an are …% between them
• text-5: the commonest ending is …% and the page must not call it a quarter without qualifying it

the ladder’s first row, wherever it is said

• reading-ladder.md has no row for Lesson 01
• …: says …% of the texts’ words after one lesson; the ladder’s own row says …%
• …: calls …% …

the index marks every briefing it calls open

• grammar/README.md says … briefings are open and marks … rows
• grammar/README.md’s row for … is …

concepts missing only as a part of speech

• a1-checklist.md: … are missing as actions and held as nouns, and the page must say so
• a1-checklist.md: … is missing and … is held, and the page must name the pair
• a1-checklist.md: … concepts appear twice — … — and one deliberate duplicate is documented

every language an etymology names must map to a family

• an etymology names a language the family map does not know, so its family is counted nowhere: …

no root but the two named may stand where only a verb can

• a root that is not a verb stands in the verb slot, which is a third root doing two jobs: …

a verb root may not stand twice in a row

• a verb root stands twice in a row, which is one root taking a second job: …

a time word in front of the subject, while it is open

• … sentences open with a time word; the shape is an open question and stood in eight when it was raised: …
• grammar/place.md must name every sentence that fronts a time word: …

a gloss may not claim a category the language lacks

• a gloss claims a category the language has not settled: …

a tense marker in front of an adjective, while it is open

• … sentences put a tense marker in front of an adjective; the shape is an open question and stood in five when it was raised: …
• grammar/tense.md must name every sentence that uses the open shape: …

the concepts the language composes instead of naming

• a1-checklist.md must name the concepts the material composes rather than names: …

the kinship block, split three ways

• a1-checklist.md: … of the people domain are missing from the index, not thirteen
• a1-checklist.md: the three parts of the kinship block are … and must account for all …

the phrasebook’s own reading level

• phrasebook.md: … of the first section’s lines need nothing past Lesson 01
• phrasebook.md does not name the lines Lesson 01 covers: …
• phrasebook.md: the first section is not complete until Lesson …
• phrasebook.md: its last word arrives in Lesson …

balance.md’s examples of a wide root

• balance.md: … roots name five families or more
• balance.md: the widest root is … at … families
• balance.md: … roots name exactly six families
• balance.md names … among the roots with six families; it has …

balance.md’s ten candidate borrowings, re-run

• balance.md’s candidate table has … rows; the paragraph above it counts …
• …
• balance.md: … of the ten candidates pass every rule
• balance.md does not name …‘s families as the dictionary has them: …
• balance.md: the roots that name a Korean source are …
• balance.md: … roots name a … language

no paragraph is printed twice

• …: a paragraph is printed twice — …
• …: a sentence is printed twice — …

subordination’s two shapes, both counted

• subordination.md’s two counts are stale; the material holds … clauses with a marker and … without

phonology.md illustrates itself

• phonology.md gives a row to … vowel sequences; … are attested
• phonology.md illustrates the sequence ’…’ with …, which is …
• phonology.md illustrates the consonant pair ’…’ with the root …, which does not contain it

proposal-sentence-types.md’s counts

• proposal-sentence-types.md: text 8 opens … sentences with yu and a verb
• proposal-sentence-types.md prints … example commands and does not say so
• proposal-sentence-types.md: the block prints a command missing from its own list of eight: …
• proposal-sentence-types.md: … of the eight are absent from the block — …

proposal-stress.md’s own lists

• proposal-stress.md no longer lists the three-syllable roots
• proposal-stress.md lists … among the three-syllable roots; they are not three-syllable roots
• proposal-stress.md names … of the … three-syllable roots, so … are unnamed
• proposal-stress.md: … roots contain an attested vowel pair, …% of the dictionary
• proposal-stress.md names roots as containing a vowel pair that do not: …
• proposal-stress.md names … of the … affected roots, so … are unnamed
• proposal-stress.md: penultimate and final differ on every root of two syllables or more, … of …, …%

pronunciation.md’s tables

• pronunciation.md gives a row to … letters; the alphabet has …
• pronunciation.md illustrates a letter with a word that is not a root: …
• pronunciation.md illustrates a letter with a word that does not contain it: …
• pronunciation.md: a vowel reduces when it is unstressed, so the rule cannot be stated as ‘no reduced variant under stress’

the A2 briefing’s theme table

• …

the figures a page states in passing

• …: says ’…’; the dictionary has …
• …: says ’…’; lessons/ holds …
• …: says ’…’; … words behave as verbs
• …: says ’…’; … roots are glossed that way
• …: says ’…’; the verb chain arrives in Lesson …
• …: says ’…’; the dictionary glosses … as …
• …: says ’…’; that spelling is …
• …: says ’…’; the ladder opens it at Lesson …

a link labelled with a number names that file

• …: a link labelled ’…’ points at …

README

• README.md says … roots; the dictionary has …

a date that has not happened

• …: … is not a date
• …: dated …, which has not happened — today is …

links

• …: broken link to …